Radar Micro-Doppler Phase Feature Extraction for Real-Time Activity Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radar-based human activity recognition methods using Micro-Doppler signatures require significant computational resources, making real-time classification on embedded platforms challenging due to complex classification algorithms and high computational demands.
Innovation Solution
A method that utilizes phase information from Micro-Doppler spectrograms to achieve high accuracy in human activity classification by converting amplitude spectrogram images to grayscale, applying binary image masks to identify Regions Of Interest (ROI), and calculating geometric and textural features, which are then classified using a trained classifier, reducing computational resources and enabling real-time implementation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex classification algorithms are used to improve human activity recognition accuracy, then classification accuracy is improved, but computational resources and device complexity increase significantly
Solution Approach 1:
The patent extracts only the essential phase information from the complete micro-Doppler spectrogram, discarding redundant amplitude and color information. By taking out only the necessary phase data and applying binary masking to identify regions of interest, the system reduces computational complexity while maintaining high recognition accuracy above 80%.
Solution Approach 2:
The patent changes the representation parameters from full-color spectrogram images to grayscale images with binary masks. This parameter transformation simplifies the data structure and reduces the computational burden on classification algorithms, enabling real-time processing on embedded platforms while preserving the essential characteristics needed for accurate human activity recognition.
2Measurement precision
If complex classification algorithms are used to improve human activity recognition accuracy, then classification accuracy is improved, but processing time and real-time capability deteriorate
Solution Approach 1:
The patent performs preliminary actions by converting the spectrogram to grayscale and applying binary masks to identify regions of interest before feeding data to the classification algorithm. This pre-processing simplifies the input data structure and reduces the computational time required for classification, enabling real-time processing while maintaining accuracy above 80%.
Solution Approach 2:
By extracting only the phase information and creating binary masks of regions of interest, the patent reduces the amount of data that needs to be processed in real-time. This extraction approach significantly decreases processing time and makes the system suitable for embedded platforms with limited computational power.
3Loss of information
If amplitude spectrogram images are processed directly, then more information is available, but computational complexity and processing requirements increase
Solution Approach 1:
The patent extracts only the phase information from the amplitude spectrogram, discarding the amplitude data. This extraction is justified because phase information contains the essential characteristics needed for human activity recognition, while amplitude information proves to be redundant. The resulting binary masked phase images require significantly fewer computational resources for processing.
Solution Approach 2:
The patent transforms the parameter space from amplitude-based representation to phase-based representation. This parameter change simplifies the data structure and reduces computational requirements, as phase information provides sufficient discriminative power for activity classification without needing to process the more complex amplitude variations.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method achieves over 80% accuracy in human activity recognition with significantly reduced computational resources, facilitating real-time classification and embedding on platforms, surpassing current techniques in both accuracy and efficiency.
Implementation Method 1
Radar is becoming increasingly irreplaceable in this field due to its unique advantages... uses the Micro-Doppler signature, which is a powerful representation of body micro-motions, synthesizing the Doppler components induced by different body parts
Data Source
Figure 1
Figure 2
Figure 3(a)~3(f)
AI summary
The invention relates to a method (100) of classifying in real-time human activity from a micro-Doppler signature image based on features extracted from phase information and unwrapped phase information. The invention relates also to a device for characterizing in real-time a human activity comprising : a radar (2) transmitting and receiving radar signals having a software interface for configuring the form of the transmitted signal and processing and calculation means (3) coupled to the radar, configured to, in real-time, implementing the method.